bura.brunel.ac.uk · Web viewThe two outputs of the model, i.e., cracking and rutting, showed...
Transcript of bura.brunel.ac.uk · Web viewThe two outputs of the model, i.e., cracking and rutting, showed...
A FUZZY INFERENCE SYSTEM BASED ASPHALT SURFACE DETERIORATION PREDICTION MODEL DUE TO COMBINED INTERACTION OF DYNAMIC LOADING-
WATER-PAVEMENT
Fauzia M Saeed, Corresponding AuthorDepartment of Civil and Environmental Engineering Brunel University, London Kingston Ln, Uxbridge, Middlesex UB8 3PH Tel: 44-745-9896-084; Email: [email protected]
Mujib M RahmanDepartment of Civil and Environmental Engineering Brunel University, London Kingston Ln, Uxbridge, Middlesex UB8 3PH Tel: +44 7894 339 752 Email: [email protected]
Maher S MahmoodCivil Engineering DepartmentUniversity Of Anbar, Ramadi, Anbar, IraqTel: ++9647736389271; Email: [email protected]
Phil CollinsDepartment of Civil and Environmental Engineering Brunel University, London Kingston Ln, Uxbridge, Middlesex UB8 3PH Tel: ++44 1895 266082; Email: [email protected]
Word count: 4571 words text + 5 Figures and 6 Tables × 250 words (each) = 7321 words
Submission Date: 30/07/2018
123456789
1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556
Saeed, Rahman, Mahmood and Collins
ABSTRACT
A Fuzzy Inference System (FIS) based deterioration prediction models has been developed in two stages. Firstly, experimental work was conducted to evaluate the performance of asphalt surfaces due to the combined action of water and dynamic loading. Then, a FIS model was developed using high dimensional inputs, such as three types of asphalt surfaces, three aggregate sizes, and two weather conditions (dry and wet), and repeated loading at two frequencies. The two outputs of the model, i.e., cracking and rutting, showed excellent agreements with the experimental measurement of cracking and rutting. The validation and sensitivity analysis were also conducted to evaluate the model performance and to evaluate the influence of each input parameters on distress prediction. The FIS models demonstrated the potential for further development as a routine prediction model to differentiate the performance of asphalt surfaces subjected to dynamic loading while submerged in water.
Key words: Deterioration prediction model, FIS, Wang & Mendel technique, fatigue cracking, rutting, fuzzy logic.
2
123456789
10111213141516171819
Saeed, Rahman, Mahmood and Collins
INTRODUCTION
Asphalt pavements are complicated physical structures that react in a complicated way to the impacts of the environmental and to the load-related variables (1). Water is one of the environmental variables behind asphalt surface damage. It is recognised that the water on the surface or water builds up in the pavement structure due to rainfall or poor drainage accelerates surface damage with repetitive traffic loading, which may subsequently occur in pavement surface layer spalling or loosening, leading to localised and structural damage (2-6). When traffic moves on the submerged pavement, the interaction of tire-water- pavement creates pore water pressure. Despite extensive studies conducted on water related material degradation in the last fifty years, research on the impact of water pressure on pavement performance caused by the dynamic action is still very limited.
To address this shortcoming, authors of this paper have developed a novel experimental method to measure water pressure under a pavement when pavement is subjected to flooding and trafficking at the same time (7, 8). The impact of load frequency, tires tread shape and thickness, and depths of surface water on pore water pressure in the pavement have been investigated in detail. The results showed that pore water pressure under the pavement depends on loading speed and shape and thickness of the tread. Water pressure increased significantly when high frequency loading is combined with square types of tread with deep tread depth when water trapped inside the groove.
Rutting, cracking and raveling are the three main distresses that can happen on asphalt surfaces when pavement is flooded with water and experiences repeated traffic loading. There are different standard laboratory tests available to assess and quantify these distresses. However, in the presence of water, these distresses can occur simultaneously. It is, therefore, essential to develop a test method that can simulate combined traffic-tire-water-pavement interaction in a controlled laboratory environment. To address this issue, the test developed by the authors to measure pore water pressure under a pavement has been used in a repeated load scenario. The combination of load magnitude, speed, tread characteristics and depth of surface water that create maximum pore water pressure, from Saeed et al (2018), were used (7, 8). By doing this, it was possible to quantify the type and amount of deterioration on different types of surfaces and compared to each other (9).
Once the deterioration has been quantified, it is important to develop a prediction model for evaluating the relative performance of different types of surfaces, which can be utilised to optimise the selection of asphalt surface suitable for specific loading and climatic conditions. A prediction model usually is done by learning from the past for which actual data is gathered and analysed to investigate the resulting pattern (10). There is a plethora of prediction models available in the literature, from simple deterministic linear regression model to a probabilistic Markov chain to artificial intelligence based (11-13). All these methods have their merits and short-comings. After a careful literature review, in this study, a Fuzzy Inference system (FIS) has been used (14). FIS modelling has excellent learning capabilities, requires less computational effort, suitable to deal with high dimensional problems and easier to implement (15). A brief overview of the method has been given in the modelling section of this paper.
OBJECTIVES
The primary objective of this study was to develop multi input models to predict deterioration of asphalt surfaces due to the combined action of repeated traffic loading with specific tire characteristics applied on submerged asphalt surfaces. The input parameters for this study were asphalt surface type, aggregate size and loading frequency. Three asphalt surfaces, namely, hot rolled asphalt (HRA), stone mastic asphalt (SMA) and porous asphalt (PA) were chosen as these are the most commonly used surfacing types. Each of these mixtures was produced with different sizes of aggregates to assess their impact on mixture performance and was tested in two environmental, dry and wet, and in two loading conditions, 5Hz and 10Hz, to simulate different loading speeds.
In the prediction models, both rutting and cracking were considered as the main distresses. After developing the model, validation exercise and sensitivity were carried out to evaluate the model accuracy and the influence of each parameter in asphalt performance.
3
12
3456789
1011
12131415161718
19202122232425262728293031323334353637383940414243
4445464748495051
525354
Saeed, Rahman, Mahmood and Collins
This study consists of two main stages, in stage 1, brief description of the experimental study on different types of asphalt surfaces are presented. Detail explanation of stage 1 research was reported in (9). In stage 2, a brief overview of the FIS system followed by model generation from stage 1 experimental data, results in analysis, and model validations and finally key conclusions, are presented.
STAGE 1: ASPHALT DISTRESS MEASUREMENT DUE TO COMBINED ACTION OF TRAFFIC-WATER-PAVEMENT
Sample preparation and test set-up
In total, 36 slabs (200mm×200mm×50mm) were manufactured using relevant BS EN standards. All compacted slabs were tested for bulk density as detailed in BS EN 12697-6: 2003(16). The actual percentage of air voids of each test specimen were calculated according to BS EN 12697-8: 2003 (17). The target void contents were 8-13% for the SMA; 4-6% for the HRA and >16% for the PA (18-20). Sample characteristics and mixture properties are given in Table 1.
TABLE 1 Specimen characteristics and mixture properties
Mixture Type
Nominal maximum size (mm)
No of sample
Specimen Size
(mm3)
Void contents (%)
Max Min Std
HRA 10 6 200×200×50 7.81 5.97 0.73614 6 200×200×50 8.00 5.20 1.0176* × × × × ×
SMA6 6 200×200×50 12.78 8.97 1.421
10 6 200×200×50 13.63 10.36 1.37114 6 200×200×50 12.50 10.33 1.024
PA14 6 200×200×50 20.89 18.00 1.315
10** × × × × ×6** × × × × ×
*not common type of road surface. ** Not suitable of PA mixture design.
The experimental program consisted of designing a novel loading plate, and an INSTRON 8501 machine to apply dynamic loading. Details test set-up can be found in (9). A picture of the specimens, the schematic diagram of the loading arrangement and the test setup are presented in Figure 1. A 5kN sinusoidal load at a frequency 5Hz and 10Hz were applied for 20,000 to 40,000 load cycles to produce significant damage to the surface (9).
Specimens ready for testing Test setup Distress measurement setup
FIGURE 1 Experimental works (stage 1)
For each mixture type, three specimens were tested in dry condition and three in wet condition. The asphalt surface was immersed with 1-2mm water, and this depth was constantly maintained by
4
123456789
10111213141516171819
2021222324252627282930
3132333435
Saeed, Rahman, Mahmood and Collins
regular feeding of water throughout the duration of the test. Specimens tested in wet condition went through overnight conditioning in water at room temperature to ensure saturation before testing. The distresses were measured at every 1000 cycles by a microscope, digital images and grid system to ensure the consistency in measuring. The result of cumulative cracking and rutting, both measured in mm, of asphalt specimens, is given in Figure 2.
0 5000 10000 15000 20000 25000 30000 35000 40000 450000
100200300400500600700800900
HRA-DSMA-WHRA-W
Cumulated load pulse
Crac
king
(mm
)
0 5000 10000 15000 20000 25000 30000 35000 40000 450000
100200300400500600700800900
HRA-D
Cumulated load pulse
Crac
king
(mm
)
a. Cumulative cracking for 14mm SMA, PA and HRA
b. Commutative cracking for 10mm SMA and HRA
0 5000 10000 15000 20000 25000 30000 35000 40000 450000
100200300400500600700800900
SMA-D
Cumulated load pulse
Crac
king
(mm
)
0 5000 10000 15000 20000 25000 30000 35000 40000 4500002468
101214161820
HRA-DSMA-WHRA-W
Cumulated load pulse
Rutti
ng (m
m)
c. Commutative cracking for 6mm SMA d. Cumulative rutting of SMA, PA and HRA14 mm
0 5000 10000 15000 20000 25000 30000 35000 40000 450000
5
10
15
20
HRA-D
Cumulated load pulse
Rutt
ing
(mm
)
02468
101214161820
SMA- D
Cumulated load pulse
Rutt
ing
(mm
)
e. Cumulative rutting of SMA and HRA10 mm f. cumulative rutting of SMA 6 mm
FIGURE 2 Cumulative of average cracking and cumulative maximum Rutting
The results showed that depending on the kind of asphalt surfaces, the presence of water accelerates cracking, rutting and other distresses such as raveling. The cracking propensity was found severe in highly open graded mixtures then the gap graded hot rolled asphalt (Figures 2a). Compared to dry condition measurement, the appearance of surface crack was around seven times faster in porous asphalt tested in wet conditions. It is interesting to note that while porous asphalt is designed to drain water within open voids, the continuous presence of water on the surface water combined with loading can significantly reduce their load-bearing capacity. It is consequently essential proper drainage for adequate performance of open graded mixtures reducing the probability of pore water build up in the clogged voids.
All tested SMA mixtures showed good rutting resistance compared to porous and hot rolled mixtures (Figures 2d-2f), although their cracking resistance was significantly decreased in the presence of water. Besides, while both 10mm and 14mm HRA presented the best performance
5
123456
789
101112131415161718192021222324
Saeed, Rahman, Mahmood and Collins
regarding resistance to cracking, the rutting was significantly higher compared to the other two mixtures. However, at the end of 40,000 pulses, the porous asphalt showed significant rutting (Figure 2d). The best performance was observed in 10mm SMA. When compared the impact of aggregate size, the gradation seemed to have more impact on load-bearing capacity than the size of aggregates. Air voids do not appear to influence wet condition performance. For instance, notwithstanding similar void contents in SMA 14 and SMA 10, SMA 10 (Figures 2a-2f) was not very sensitive to wet conditions as it was in SMA14. It appeared that aggregate nominal size may influence wet condition performance.
STAGE 2: DETERIORATION PREDICTION MODEL BY FIS
Fuzzy interface system (FIS) is one of the most popular methods used in classification and prediction problems (21). Fuzzy inference is a technique that interprets the values in the input vector and, based on user-defined rules, assigns values to the output vector (22). Dehzangi et al., (2007) stated the benefits of this method that is represented by the knowledge in the form of If-Then rules, interpreting the mechanism of logic in human-understandable terms. Fuzzy logic has advantages over other computational techniques as it can take linguistic information from human experts and combines it with numerical data. In addition, it can approximate complex nonlinear functions with simpler models (23). Fuzzy based model is developed in two main steps, generation of membership functions and fuzzy rules. A short description of each of these steps is given below.
Membership functions generationWithin the fuzzy approach, the fuzzy set A of universe X is determined by the function μA(x), named the membership function of set A (24).
µA: ×→ [0, 1]
Where μA(x) = 1 if x is totally in A; μA(x) = 0 if x is not in A; 0 < μA(x) < 1 if x is partly in A. in this research, the membership functions of inputs variables are generated by data clustering method.
Data clusteringNumerical data clustering is the foundation of various modelling and classification algorithms to evaluate their performance (25). It separates the data set into many data subsets, such that the similarity within a subset is higher than between the subsets. A similarity among elements of input vectors is an essential feature to achieve data clustering (26). The most common clustering method, one has been used in this study, is k-means clustering. The fundamental thought of this clustering technique is to choose k initial cluster means, or centres randomly. After many repetitions, certain initial cluster means are updated in such a way that they represent the data clusters as much as possible (27). A limitation of the k-means clustering algorithm is that the number of clusters is fixed; after k is chosen, there will always be k cluster means or centres (21). The k-means algorithm can avoid this difficulty by eliminating the excess clusters. A cluster center may be removed if it does not have enough samples it is likely to prevent this problem by picking a large enough k (26). The steps for k-mean clustering technique are as follows;
i. Initialise Ci by randomly choosing C points from among all the data points.ii. Compute the membership matrix (U), where the element (uij) is 1 if the jth data point xj
belongs to the group i and 0 oppositely.iii. Compute the fitness function by using the following equation. Stop if the fitness function
value is lower than a certain threshold value:
J=∑i=1
c
¿ Ji=∑i=1
c
(∑ k , xk¿∈ ci‖ xk−c i ‖2)¿
6
123456789
10111213141516171819202122232425262728293031323334353637383940414243444546474849505152
53
Saeed, Rahman, Mahmood and Collins
Update the cluster center Ci and calculate the new matrix (U). The k-means clustering algorithm is iterative. Accordingly, it is hard to forecast its convergence to the best solution (21).
Fuzzy rules
A rule containing several fuzzy If-Then rules (28), and they are 1) a database which defines the membership functions in the fuzzy sets used in the fuzzy rules; 2) a decision-making unit which performs the inference operations on the rules; 3) fuzzification interface which transforms the crisp inputs into degrees of a match with linguistic values; 4) defuzzification interfaces which transform the fuzzy results of the inference into a crisp output. The number of rules of a complete rule set is equal to
Π in=1mi
Where m is the number of membership functions for input i, and n is the number of inputs
The fuzzy rules are generated either from skilful experience or numerical data (29). In this study, the widely used Wang & Mendel technique was adopted to create fuzzy rules mechanically from numerical data (30). This method needs predefined fuzzy membership functions for each input and output (21, 31). It begins by performing one rule for each data pair of the training set The ith pair rule is as follow:
IF x1 IS A1i AND x2 is A2
i …. AND xp THEN y is C i
The fuzzy sets A1i are those for which the degree of match of X j
i is maximum for each input variable j from pair i. The fuzzy set 𝐶𝑖 is the one for which the degree of match of the observed output, y, is maximum (32).
DETERIORATION PREDICTION MODEL
Figure 3 shows the flowchart of different stages in the damage prediction model. A short description of each of these stages is given below.
Note: 8SQ refers to 8mm deep square tread
FIGURE 3 Flowchart for prediction model development.Data Manipulation
7
123456789
10111213141516171819202122232425262728293031323334
353637383940
Saeed, Rahman, Mahmood and Collins
For building a pavement deterioration, the three asphalt types, three aggregate sizes, two weather conditions, one load and two frequencies were used as FIS inputs, and a measured cracking and rutting were defined as the FIS output. Cracking and rutting were used as an output parameter in the first model and the second model respectively. It was assumed that the void content remains constant even after the progressive damage of the asphalt surfaces during testing.
The severity level of each distress type, as shown in Table 2, was determined by using the Distress Identification Manual for the Long-Term Pavement Performance Program (33). All linguistic data information was then converted to the numerical numbers. For example, each mixture was given a numerical identification number, such as HRA=1, SMA=2, PA=3, and for aggregate size, 14 was for a large stone, 10 for medium size stone and 6 for small size stone. The dry condition was referred to as “0” while the wet condition was given number “1”.
Membership function
The membership functions of input variables were generated by k-means clustering using Fuzzy Inference System Professional (FisPro) software (34). As both rutting and cracking occur in asphalt surface simultaneously, the membership functions for inputs variables of both rutting and cracking in the deterioration model were kept the same. For each parameter type, three triangular membership functions representing the range of variability (low, medium, and high), as shown in Table 2, were generated. The seven triangular membership functions of output were generated manually. The membership functions are shown in Figure 4.
TABLE 2a Criteria used to evaluate the severity levels of cracking
Mixture ID
Severity level Low
Severity level Medium Severity level High
70 mm to 100mm with a few
connecting cracks
100mm to 150mm with interconnected
cracks
> 150mm interconnected cracking forming a complete pattern and pieces
move with loadingHRA10-D 9320 10236 14664HRA10-W 7180 7820 15096HRA14-D 40,000* - -HRA14-W 40,000* - -SMA6-D 4350 5120 6420SMA6-W 4313 4780 5640SMA10-D 4785 5192 6448SMA10-W 3290 3712 4576SMA14-D 6037 7360 10001SMA14-W 3732 4344 5420
PA14-D 5025 5400 5960PA14-W 1110 1432 2000
TABLE 2 b Criteria used to evaluate the severity levels of rutting
8
123456789
1011121314151617181920212223242526
28293031323334353637383940
Saeed, Rahman, Mahmood and Collins
Mixture ID
DMRB criteria Low Severity <6mm; 6mm< Medium <11mm; High severity>11mm
@ 20,000 load cycles @ 40,000load cyclesMeasured rutting
(mm) Severity level Measured rutting (mm) Severity level
HRA10_D 11.4 H - -HRA10_W 13.2 H - -HRA14_D 14.3 H 14.5 HHRA14_W 17.5 H 18 HSMA6_D 8.3 M - -SMA6_W 9.5 M - -SMA10_D 7.5 M - -SMA10_W 9.4 M - -SMA14_D 4.5 M 9.9 MSMA14_W 9.0 M 13 H
PA14_D 8.2 M 13 HPA14_W 9.5 M 18 H
* The cracks length in HRA14 mixture was 10 mm for both cases with the number of load pulse 40,000
Aggregate sizes MFs Void contents MFs
Asphalt type MFs Cumulated load pulse MFs
Rutting MFs Cracking MFs
FIGURE 4 Membership functions for model developement
Fuzzy Rule
9
1
23456
789
1011
Saeed, Rahman, Mahmood and Collins
The FisPro software was used for automatic generation of fuzzy rules from the numerical data. The generation of fuzzy rules of the deterioration model reported in this study was challenging and complicated as it consisted of six inputs and one output for each model. To overcome the problem of generation of the fuzzy rules and membership functions with a high-dimensional problem, the membership functions of inputs parameters were created based on the k-means clustering technique in (FisPro) software (21). FisPro offers the possibility to generate fuzzy inference systems and to use them for reasoning purposes, especially for simulating a physical or biological system (31). In total 158 rules were created for each model using the logic given in the fuzzy inference system section of this paper. A typical example is given in the following Table 3.
TABLE 3 Fuzzy If-Then rules generated by cracking (selected)
Rules
Input Rule - If “Type if Asphalt ” is … and “Size if Aggregate” is … Output Rule- The Cracking Severity Level is …
Asphalt Type
Agg
rega
te
Size
Wea
ther
Freq
uenc
y
Void Content No of Cycles
1 SMA Large Dry 5Hz Medium 0-6000 Low2 PA Large Dry 5Hz Very high 0-6000 Low3 HRA Medium Dry 5Hz Medium 0-6000 Low4 HRA Large Dry 5Hz Medium 0-6000 Low
,,
157 HRA Large Wet 5Hz Medium 34000-40000 Low158 SMA Medium Wet 10 Hz High 34000-40000 High
RESULTS AND DISCUSSION
As mentioned earlier, two separate deterioration models for cracking and rutting were developed. For each model, six input variables (asphalt type, aggregate sizes, weather conditions, frequencies, void contents and cycles numbers) were used to generate output for rutting and cracking. The results are given below.
Pavement Deteriorations (Rutting and Cracking)
Figure 5a and Figure 5b show model correlation for rutting and cracking respectively. It can be seen that a correlation of approximately 95.8% was achieved between measured and predicted rutting and 98% correlation between measured and predicted cracking. This indicates a very good accuracy of both models.
The performance of a fuzzy inference system-based cracking
b) The performance of a fuzzy inference system based on rutting
FIGURE 5 The performance of a fuzzy inference system
10
123456789
10111213
1415
161718192021222324252627282930
31
Saeed, Rahman, Mahmood and Collins
Table 4 shows the coefficient of determination together with the root mean square error (RMSE) and mean absolute error (MAE) to determine the level of agreement of the rutting and cracking values in the two data sets. The error levels are less than one for rutting, but high in the cracking. This is because the measurement of rutting was only up to 20mm and it was concentrated in confined areas under the loading. On the other hand, cracking was distributed across the slab and measured up to 800mm after the test which would create some variability. Despite this, high correlations in both cases indicate good model accuracy.
TABLE 4 :Model performance of asphalt deteriorations
Model Validation
The entire data set used to develop the prediction model were randomly split into training data (around 80%) and testing data (around 20%). The training data were used to generate the prediction models while the testing data were used to validate the models. The data for both models were split by using SSPS software(35). SPSS can automatise this selection process without requiring multiple steps. The process to split training data and testing data have been repeated to run models five times for both cracking and rutting to ensure all data were included.
The estimated root mean square errors (RMSE) for the models were used to compare their efficiency in prediction. Results for cracking and rutting are given in Tables 5a and 5b respectively.
TABLE 5a The validation of model performance for cracking
Model No Training data 80% of all data Testing data 20% of all dataR2 (%) RMSE MAE R2 (%) RMSE MAE
1 97.9 30.174 22.129 97.9 34.356 24.9942 98.1 30.305 21.96 98.3 27.891 21.1463 97.9 31.513 22.537 98.2 29.834 22.6144 98.2 29.919 22.152 97.7 30.315 21.3245 98.3 28.093 20.819 98.2 28.417 21.678
TABLE 5b The validation of model performance for rutting
The model developed for cracking has better accuracy (R2= 97.9 to 98.3) than the rutting model (R2= 85.1 to 96.9). Despite this difference, both models have shown very good agreement with the measured data.
Sensitivity analysis
11
Deteriorations Types R2 (%) RMSE MAERutting 95.8 0.995 0.686
Cracking 98 27.45 19.808
Model No Training data 80% of all data Testing data 20% of all dataR2 (%) RMSE MAE R2 (%) RMSE MAE
1 95.8 0.882 0.601 95.4 0.846 0.5882 85.4 2.808 2.293 96.4 0.721 0.553 96.9 0.726 0.54 95.7 0.842 0.614 96.2 0.795 0.587 96.8 0.7 0.5815 88.1 1.457 0.971 95.6 0.878 0.641
123456789
101112131415161718192021222324252627282930
31323334353637383940414243444546474849
Saeed, Rahman, Mahmood and Collins
To study the influence of each input variable in the performance of the asphalt surface, a sensitivity analysis was conducted to explore the influence of each input variables on the efficiency of the deterioration prediction models. For instance, the sensitivity investigation was conducted by generating the FIS model by respecting the effect of individual input and inactive effects of other inputs.
The correlation between the fuzzified and each input variable are shown in Table 6. It can be seen clearly that the determination coefficients for asphalt type, aggregate size, weather conditions, frequency, void contents and the number of cycles were 15.1, 3.7, 10.8, 27.7, 0.1 and 28.6 for rutting and 0.1, 1, 10.1, 27.1, 39.4 and 7.4 for cracking. It was noticed that for this data set, the influence of void contents is significantly higher than that of other parameters on cracking; and in rutting number of cycles has a higher impact. In addition, as these variables are related to each other, the combined impact will be higher.
TABLE 6 Sensitivity level for each input variable
Input VariableDeterioration modelRutting CrackingR2 (%) R2 (%)
Asphalt Type 15.1 0.1Aggregate Size 3.7 1Weather Conditions 10.8 10.1Frequency 27.4 27.1Void Contents 0.1 39.4Number of Cycles 28.6 7.4
CONCLUSION AND RECOMMENDATIONS
Multi input fuzzy-based deterioration prediction models have been developed to evaluate the combined interaction of traffic loading and water on asphalt surfaces. The prediction accuracy of the model was approximately 99% with the experimentally measured distresses such as cracking and rutting. The accuracy of rutting prediction model was marginally better than cracking prediction model. This was due to the spread of distresses on the tested specimens. The validation of the models also accurately predicted both distresses from a randomly selected dataset.
The sensitivity analysis to evaluate the influence of each variable on the model performance showed that irrespective of mixture type, mixture parameters (aggregate size, void contents), traffic parameters (loading frequency) and environmental factors (wet condition) have an impact on either cracking or on rutting or on both. Further development could be extended to implement this model on independent dataset experiences similar traffic and environmental loading.
12
123456789
1011121314151617
181920212223242526272829303132333435363738394041424344454647
Saeed, Rahman, Mahmood and Collins
REFERENCE
1. Saeed, F., S. Qamariatul, M. Rahman, and A. Woodside. The State of Pothole Management in UK Local Authority. Bituminous Mixtures and Pavements VI, 2015, pp. 153-159.
2. Lindly, J. K., and A. S. Elsayed. Estimating Permeability of Asphalt-Treated Bases. Transportation Research Record, No. 1504, 1995, pp. 103-111.
3. Karlson, T. K. Evaluation of Cyclic Pore Pressure Induced Moisture Damage in Asphalt Pavement. University of Florida, 2005.
4. Kim, Y., J. S. Lutif, A. Bhasin, and D. N. Little. Evaluation of Moisture Damage Mechanisms and Effects of Hydrated Lime in Asphalt Mixtures through Measurements of Mixture Component Properties and Performance Testing. Journal of Materials in Civil Engineering, Vol. 20, No. 10, 2008, pp. 659-667.
5. Willway, T., L. Baldachin, S. Reeves, and M. Harding. The Effects of Climate Change on Highway Pavements and how to Minimise them: Technical Report. The Effects of Climate Change on Highway Pavements and how to Minimise them: Technical Report, Vol. 1, No. 1, 2008, pp. 1-111.
6. Kennedy, T. W., F. L. Roberts, and K. W. Lee. Evaluation of Moisture Effects on Asphalt Concrete Mixtures, Transportation Research Record, No. 911, 1983, pp. 134-143.
7. Saeed, f., M. Rahman, and D. Chamberlain. A Novel Laboratory Test Method to Measure Dynamic Water Pressure Underneath a Cracked Concrete Pavement. . IJPEAT the International Journal of Pavement Engineering and Asphalt Technology, Vol. 19, No. 2, 2018.
8. Saeed, F., M. Rahman, and D. and Chamberlain. Impact of Tire and Traffic Parameters on Water Pressure in Pavement. Journal of Transportation Engineering, Part B: Pavements (ASCE).., 2018.
9. Saeed, F., M. Rahman, and D. Chamberlain. A Novel Test Method to Evaluate the Performance of Asphalt Surface due to Combined Action of Water and Dynamic Loading. Construction & Building Materials, 2018.
10. Vaidehi, V., S. Monica, S. Mohamed Sheik Safeer, M. Deepika, and S. Sangeetha. A Prediction System Based on Fuzzy Logic. , 2008.
11. Abaza, K. A. Back-Calculation of Transition Probabilities for Markovian-Based Pavement Performance Prediction Models. International Journal of Pavement Engineering, Vol. 17, No. 3, 2016, pp. 253-264.
12. Saleh, S. E., G. J. Awda, and N. G. Ahmed. Development of Pavement Condition Index Model for Flexible Pavement in Baghdad City. Journal of Engineering, Vol. 14, No. 1, 2008, pp. 2120-2135.
13. 1 Al-Mansour, A. I., K. C. Sinha, and T. Kuczek. Effects of Routine Maintenance on Flexible Pavement Condition. Journal of Transportation Engineering, Vol. 120, No. 1, 1994, pp. 65-73.
14. Mahmood, M., M. Rahman, and S. Mathavan. A Multi-Input Deterioration-Prediction Model for Asphalt Road Networks. In Proceedings of the Institution of Civil Engineers-Transport, Thomas Telford Ltd, 2018, pp. 1-12.
15. Sheta, A. Software Effort Estimation and Stock Market Prediction using Takagi-Sugeno Fuzzy Models. In Fuzzy Systems, 2006 IEEE International Conference On, IEEE, 2006, pp. 171-178.
16. British Standards Institution. BS EN 12697-6: 2003: Bituminous mixtures. Test methods for hot mix asphalt. Determination of bulk density of bituminous specimens. British Standards Institution, London, 2003.
13
123456789
101112131415161718192021222324252627282930313233343536373839404142434445464748
49
Saeed, Rahman, Mahmood and Collins
17. British Standards Institution. BS EN 12697-8: 2003: Bituminous mixtures. Test methods for hot mix asphalt. Determination of void characteristics of bituminous specimens. British Standards Institution, London, 2003.
18. British Standards Institution. BS EN 13108-5:2016: Bituminous mixtures. Material specifications. Stone Mastic Asphalt. British Standards Institution, London, 2016.
19. British Standards Institution. BS EN 13108-4:2016: Bituminous mixtures. Material specifications. Hot Rolled Asphalt. British Standards Institution, London, 2016.
20. British Standards Institution. BS EN 13108-7:2006: Bituminous mixtures. Material specifications. Porous Asphalt. British Standards Institution, London, 2006.
21. Mahmood, M. S. Network-Level Maintenance Decisions for Flexible Pavement using a Soft Computing-Based Framework, 2015.
22. Borkar, A. D., and M. Atulkar. Fuzzy Inference System for Image Processing. International Journal of Advanced Research in Computer Engineering & Technology (IJARCET), Vol. 2, No. 3, 2013, pp. pp: 1007-1010.
23. Dehzangi, O., M. J. Zolghadri, S. Taheri, and S. M. Fakhrahmad. Efficient Fuzzy Rule Generation: A New Approach using Data Mining Principles and Rule Weighting. In Fuzzy Systems and Knowledge Discovery, 2007. FSKD 2007. Fourth International Conference On, IEEE, 2007, pp. 134-139.
24. Negnevitsky, M. Artificial Intelligence: A Guide to Intelligent Systems. Addison Wesley., 2002.
25. Jain, A. K. Data Clustering: 50 Years Beyond K-Means. Pattern Recognition Letters, Vol. 31, No. 8, 2010, pp. 651-666.
26. Naik, V. C. Fuzzy C-Means Clustering Approach to Design a Warehouse Layout. , 2004.27. Christiansen, B. Atmospheric Circulation Regimes: Can Cluster Analysis Provide the
Number? Journal of Climate, Vol. 20, No. 10, 2007, pp. 2229-2250.28. Jang, J. ANFIS: Adaptive-Network-Based Fuzzy Inference System. IEEE Transactions on
Systems, Man, and Cybernetics, Vol. 23, No. 3, 1993, pp. 665-685.29. Nelles, O., M. Fischer, and B. Muller. Fuzzy Rule Extraction by a Genetic Algorithm and
Constrained Nonlinear Optimization of Membership Functions. In Fuzzy Systems, 1996., Proceedings of the Fifth IEEE International Conference On, IEEE, 1996, pp. 213-219.
30. Wang, L., and J. M. Mendel. Generating Fuzzy Rules by Learning from Examples. IEEE Transactions on Systems, Man, and Cybernetics, Vol. 22, No. 6, 1992, pp. 1414-1427.
31. Chen, S., and Y. Chen. Automatically Constructing Membership Functions and Generating Fuzzy Rules using Genetic Algorithms. Cybernetics &Systems, Vol. 33, No. 8, 2002, pp. 841-862.
32. Guillaume, S., Charnomordic, B.,Lablee, J.,. FisPro: An Open Source Portable Software for Fuzzy Inference Systems. , 2013.
33. Miller, J. S., and W. Y. Bellinger. Distress Identification Manual for the Long-Term Pavement Performance Program, 2003.
34. Guillaume, S., and B. Charnomordic. Learning Interpretable Fuzzy Inference Systems with FisPro. Information Sciences, Vol. 181, No. 20, 2011, pp. 4409-4427.
35. Landau, S. A Handbook of Statistical Analyses using SPSS. CRC, 2004.
14
123456789
101112131415161718192021222324252627282930313233343536373839404142
434445464748